AI-first multi-agent platform foundation with FastAPI backend, React frontend, and plugin architecture for building multi-tenant AI applications. Includes MCP server support for agent tool execution.
10 tools with basic schemas and descriptions, but significant gaps in parameter documentation and error handling. Most tools lack descriptions for individual parameters (pattern:tool-description violation). The code_execute tool has a good description but no input validation guidance. rag_search is well-described but lacks output schema documentation. No tool declares permissions or error recovery paths. Naming is generally clear (verb_noun pattern), but several tools lack depth in their descriptions (under 100 chars). No evidence of idempotency hints, confirmation patterns for destructive ops, or structured error responses.
Run an AI completion using the platform's AI gateway
List available AI models
Execute Python code in a secure sandbox with resource limits. Use this to perform calculations, data processing, or run algorithms. The sandbox blocks network access, filesystem writes, and dangerous imports.
Echo back the input — useful for verifying agent tool wiring.
List files in storage
Upload a file to storage
Parameter descriptions missing or minimal. 'ai_list_models', 'file_list', 'job_list' have empty input schemas with no parameter guidance. LLMs cannot infer parameter meaning from names alone.
No output schema documentation. Tools return results but LLMs have no visibility into response structure (fields, types, pagination). Breaks downstream tool chaining.
Destructive/irreversible tools (file_upload, job_create, webhook_create, code_execute) lack confirmation patterns or dry-run support. No guidance on error recovery or rollback.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 54 | 2026-07-28+ | v2 |
Create a background job
List background jobs
Search documents and agent memory using AI-powered hybrid search. Combines vector similarity (semantic meaning) with full-text keyword matching. Use this to find relevant information across uploaded documents, JSON data, PDFs, spreadsheets, and agent memory. Supports filtering by source type, date range, and tags.
Create a webhook endpoint
No permission declarations or scope hints. Tools do not state what authorization level they require (read:files, write:jobs, etc.), preventing least-privilege agent configuration.
code_execute tool accepts arbitrary Python code with minimal validation guidance. Description mentions 'secure sandbox' but no details on what is blocked, timeout behavior, or error messages for policy violations.